UC Santa Cruz systems researcher: semantic profiling and benchmarking for AI agents and network experiments
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Output metrics are not yet scored: the OpenAlex snapshot reports an h-index and i10-index of zero with no usable multi-year publication series, so this axis — total works, authorship position, and venue spread — cannot be responsibly quantified.
Impact—
Impact cannot be scored yet: with an h-index of 0 and an i10-index of 0, the record lacks sufficient citing works to measure citation-based influence. This axis would normally capture citations, citing works, and uptake of the 2026 arXiv outputs.
Focus0/100
The focus score is 0, indicating minimal topical concentration. The OpenAlex record spreads across five topics — Security and Verification in Computing (10 works), Parallel Computing and Optimization Techniques (9), Logic, programming, and type systems (6), Distributed systems and fault tolerance (5), and Access Control and Trust (4) — with the top two nearly tied, while fieldDist splits 0.5/0.5 between networking and agent observability. This breadth reads as an early-career agenda still consolidating rather than an error.
Collaboration55/100
The collaboration score is 55, mid-range. The OpenAlex record lists 15 distinct coauthors, and Evolve identifies 12 frequent collaborators — including Tianzhu Zhang, Weichen Tao, Changgang Zheng, and Meikang Qiu — pointing to steady team-based systems work. On the two reviewed 2026 papers, the researcher is first author on AgentPProf and a middle author on the NetArtifactBench study, consistent with active participation inside a working group.
Activity87/100
The activity score is 87, high on this scale. The record shows 19 works in the most recent three-year window, with countsByYear attributing all 19 to 2026, and both reviewed papers appeared as 2026 arXiv preprints. This concentrated burst is typical of an early-stage researcher whose output is recent and accumulating quickly, though the record is too new to demonstrate a sustained multi-year publishing cadence.
Reproducibility38/100
The reproducibility score is 38 (reproMean 38, computed over 2 papers reviewed by Evolve). AgentPProf is rated medium readiness: its reported 0.764 B3 F1 rests on case studies and a proprietary segmenter LLM. The NetArtifactBench paper is rated low: the 65.3% pass rate depends on specific model access for scoring. Releasing the segmenter and evaluation harness would be the clearest path to a higher score.
Research summary
Yusheng Zheng is a computer scientist at the University of California, Santa Cruz whose profile, resolved from OpenAlex bibliographic data (ORCID 0009-0004-3193-4356), centers on AI-agent systems and network-systems observability. The record is early-stage but fast-moving: 19 works in the most recent three-year window, all timestamped 2026 in countsByYear, yielding an activity score of 87. Indexed work spans five OpenAlex topics — led by Security and Verification in Computing (10 works) and Parallel Computing and Optimization Techniques (9) — and the field distribution splits evenly (0.5/0.5) between computer networking and AI-agent systems. Two 2026 arXiv papers anchor the Evolve catalog. AgentPProf (first author) proposes a semantic profiler for long-horizon AI agents, reporting 0.764 B3 F1 in trajectory segmentation and gains in fault localization; Evolve judges the gains plausible but notes the evidence rests on case studies and a proprietary segmenter LLM. The second paper (middle author) introduces NetArtifactBench, a 52-instance benchmark on which AI agents repair network-experiment artifacts, passing 65.3% overall but under 30% on cross-artifact repairs. Reproducibility readiness is mixed (mean 38 across 2 reviewed papers). Collaboration is mid-range (score 55) with 15 distinct coauthors; output and impact axes remain unscored because citation indicators currently stand at zero.
Collaboration pattern
Collaboration is mid-range (score 55) on a record of 15 distinct coauthors. Twelve recurring collaborators anchor the pattern — Tianzhu Zhang, Weichen Tao, Changgang Zheng, Long Chen, Xiaoyi Fan, Meikang Qiu, Chaokun Chang, Yu Mao, Tianyuan Wu, Yuxi Huang, Tao Ma, and Wenan Mao — indicating team-based systems work rather than solo authorship. On the two reviewed 2026 papers the researcher holds first-author position on AgentPProf and a middle position on the NetArtifactBench study.
Research keywords
Highlights
AgentPProf: semantic profiling for long-horizon AI agents
NetArtifactBench: can AI agents repair drifting network artifacts?
Timeline
- 2026Published first-author arXiv preprint AgentPProf, a semantic profiler for long-horizon AI agents reporting 0.764 B3 F1 trajectory segmentation and improved fault localization.
- 2026Co-authored the NetArtifactBench paper (52 instances), showing AI agents pass 65.3% of network-artifact repair tasks but under 30% on cross-artifact repairs.
- 2026OpenAlex countsByYear attributes all 19 works of the recent three-year window to 2026, alongside 15 distinct coauthors in the same record.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric indicators and topic counts derive from OpenAlex only; h-index (0) and i10-index (0) reflect the snapshot and are not independently verified.
- Reproducibility (reproMean 38) rests on a small sample of 2 Evolve-reviewed papers; broader output is not yet assessed.
- countsByYear covers only 2026, so multi-year productivity and trend lines cannot be computed.
- Identity is resolved from public bibliographic data linked to ORCID 0009-0004-3193-4356; no biographical details beyond institutional affiliation are asserted.